boot2

commit 3d4a605868477322720da30bfb1ce9e6abbcecb1
parent bcb244c97bd9a962f8f7fb969f770d7e8d00a195
Author: Ryan Sepassi <rsepassi@gmail.com>
Date:   Wed, 29 Apr 2026 23:22:43 -0700

aarch64: support large ADD/SUB immediates in p1_addi and p1_enter

aa64_add_imm / aa64_sub_imm encode a 12-bit immediate and silently
truncate values > 4095. Functions with large stack frames (e.g.
cc__next_nomacro1) produced a frame smaller than requested, corrupting
later stack-slot accesses.

Add aa64_{add,sub}_imm_lsl12 (shift-12 variants) and
aa64_{add,sub}_imm_any that dispatch to lsl12, plain, or lsl12+plain
depending on the immediate's magnitude. Wire p1_addi and %p1_enter
through the _any helpers.

Regression: tests/p1/large-addi.P1pp.

Diffstat:
M
P1/P1-aarch64.M1pp
 | 
34
++++++++++++++++++++++++++++++----
A
tests/P1/large-addi.P1pp
 | 
37
+++++++++++++++++++++++++++++++++++++
A
tests/P1/large-addi.expected
 | 
1
+

3 files changed, 68 insertions(+), 4 deletions(-)
diff --git a/P1/P1-aarch64.M1pp b/P1/P1-aarch64.M1pp
@@ -161,6 +161,32 @@
 %((| 0xD1000000 (<< (& imm12 0xFFF) 10) (<< %aa64_reg(ra) 5) %aa64_reg(rd)))
 %endm
 
+%macro aa64_add_imm_lsl12(rd, ra, imm12)
+%((| 0x91400000 (<< (& imm12 0xFFF) 10) (<< %aa64_reg(ra) 5) %aa64_reg(rd)))
+%endm
+
+%macro aa64_sub_imm_lsl12(rd, ra, imm12)
+%((| 0xD1400000 (<< (& imm12 0xFFF) 10) (<< %aa64_reg(ra) 5) %aa64_reg(rd)))
+%endm
+
+%macro aa64_add_imm_any(rd, ra, imm)
+%select((<= imm 4095),
+    %aa64_add_imm(rd, ra, imm),
+    %select((= (& imm 0xFFF) 0),
+        %aa64_add_imm_lsl12(rd, ra, (>> imm 12)),
+        %aa64_add_imm_lsl12(rd, ra, (>> imm 12))
+        %aa64_add_imm(rd, rd, (& imm 0xFFF))))
+%endm
+
+%macro aa64_sub_imm_any(rd, ra, imm)
+%select((<= imm 4095),
+    %aa64_sub_imm(rd, ra, imm),
+    %select((= (& imm 0xFFF) 0),
+        %aa64_sub_imm_lsl12(rd, ra, (>> imm 12)),
+        %aa64_sub_imm_lsl12(rd, ra, (>> imm 12))
+        %aa64_sub_imm(rd, rd, (& imm 0xFFF))))
+%endm
+
 %macro aa64_mov_rr(dst, src)
 %select((= %aa64_is_sp(dst) 1),
     %aa64_add_imm(sp, src, 0),
@@ -379,8 +405,8 @@ $(imm)
 
 %macro p1_addi(rd, ra, imm)
 %select((>= imm 0),
-    %aa64_add_imm(rd, ra, imm),
-    %aa64_sub_imm(rd, ra, (- 0 imm)))
+    %aa64_add_imm_any(rd, ra, imm),
+    %aa64_sub_imm_any(rd, ra, (- 0 imm)))
 %endm
 
 %macro p1_logi_ANDI(rd, ra, imm)
@@ -498,9 +524,9 @@ $(imm)
 %endm
 
 %macro p1_enter(size)
-%aa64_sub_imm(sp, sp, (& (+ (+ 16 size) 15) -16))
+%aa64_sub_imm_any(sp, sp, (& (+ (+ 16 size) 15) -16))
 %aa64_mem(ST, lr, sp, 0)
-%aa64_add_imm(x8, sp, (& (+ (+ 16 size) 15) -16))
+%aa64_add_imm_any(x8, sp, (& (+ (+ 16 size) 15) -16))
 %aa64_mem(ST, x8, sp, 8)
 %endm
 
diff --git a/tests/P1/large-addi.P1pp b/tests/P1/large-addi.P1pp
@@ -0,0 +1,37 @@
+:p1_main
+    %enter(0)
+    %li(t0, 0)
+    %addi(t0, t0, 5000)
+    %li(t1, 5000)
+    %beq(t0, t1, &ok)
+    %la(a0, &bad_msg)
+    %li(a1, 4)
+    %call(&write_stdout)
+    %li(a0, 1)
+    %eret
+
+:ok
+    %la(a0, &ok_msg)
+    %li(a1, 3)
+    %call(&write_stdout)
+    %li(a0, 0)
+    %eret
+
+:write_stdout
+    %enter(0)
+    %mov(a2, a0)
+    %mov(a3, a1)
+    %li(a0, %sys_write)
+    %li(a1, 1)
+    %syscall
+    %eret
+
+:ok_msg
+"OK
+"
+
+:bad_msg
+"BAD
+"
+
+:ELF_end
diff --git a/tests/P1/large-addi.expected b/tests/P1/large-addi.expected
@@ -0,0 +1 @@
+OK